[Gm2] Re: ISO M2 CAST faulty behaviour and more about opaque types

[Gaius Mulley]

Iztok writes:

> Gaius !
> 
> TYPE Foo = RECORD .... END;
>           pFoo = POINTER TO Foo;
>  VAR a : SYSTEM.ADDRESS;
>          pfoo : pFoo;
> 
> BEGIN
>   pfoo := SYSTEM.CAST(pFoo, a); (* fails with "arguments to CAST must
> be (Type, Variable or Constant)" *)
> END
> 
> 
> One more thing:
> 
> We have been discussing about opaque types recently and I have noticed
> the questionable implementation of the opaue types. Namely, you have
> said that you equaled them  to SYSTEM.ADDRESS. I am concerned about
> the correct type checking now. Consider situation:
> 
> DEFINITION 1
> 
> TYPE Opaque1;
> 
> IMPLEMENTATION 1
> 
> TYPE OpaqueImpl1 = RECORD ..... END;
> TYPE Opaque1 = POINTER TO OpaqueImpl1;
> 
> PROCEDURE p1 (arg1 : Opaque1);
> 
> DEFINITION 2
> 
> TYPE Opaque2;
> 
> IMPLEMENTATION 2
> 
> TYPE OpaqueImpl2 = RECORD ..... END;
> TYPE Opaque2 = POINTER TO OpaqueImpl2;
> 
> PROCEDURE p2 (arg2 : Opaque2);
> 
> 
> MODULE 3
> 
> VAR foo1 : 1.Opaque1;
>          foo2 : 2.Opaque.2;
> 
> 
> You allow: foo1 := foo2 which could be correct considering their
> primary behaviour - being references. Considering their referenced
> type it should issue at least warning because you could by this logic
> do also :
> 
> 2.p2( foo1 )
> 
> which is dangerous.
> 
> I think that the variable of typed pointer could be mutually
> assignable with another of SYSTEM.ADDRESS but not with another of
> different typed pointer, except using SYSTEM.CAST which transfers the
> reponsibility for functional disasters from compiler to developer in
> charge.
> 
> It also means that opaque types can not be equalized to SYSTEM.ADDRESS
> but considered as typed pointers of different kinds
> 
> 
> It is not fatal but needs consideration.
> 
> Anybody else with an opinion ?

Hi Iztok,

I believe these bugs have all been fixed,
[apologies if I've emailed this before, I'll scanning back through the
 bug reports]

regards,
Gaius